Kelp forest extent mapping for emerging blue-carbon accounting
Kelp forests are a contested blue-carbon sink with no approved credit methodology. Satellite remote sensing can map surface-canopy extent reliably in clear water, providing the spatial baseline that any future MRV framework will need.
Sensors
- Sentinel-2 MSI: 10 m resolution in the red and near-infrared bands (B4, B8), 20 m in the red-edge (B5, B6, B7). Five-day revisit at mid-latitudes with two satellites. The red-edge bands are particularly useful for separating floating kelp canopy from turbid water and bare rock. Free and open archive from 2015.
- Landsat 8/9 OLI: 30 m resolution across visible, NIR and SWIR bands. Sixteen-day single-satellite revisit, eight days with both. Coarser than Sentinel-2 for patchy kelp beds but the archive extends to 1984 (Landsat 5 TM), making it the only option for multi-decade baseline reconstruction.
- Planet SuperDove: 3 m resolution, eight spectral bands including red-edge at 705 nm. Near-daily revisit over most coastal areas. Well-suited to detecting small or fragmented kelp patches that fall below Sentinel-2's detection floor, though the archive depth is shorter and access requires a commercial licence.
- WorldView-3 (coastal and NIR bands): 0.31 m panchromatic, 1.24 m multispectral including a dedicated coastal-blue band (400–450 nm) and four NIR bands. Useful for validating canopy classifications at sub-metre detail, but tasked on demand rather than systematic, making time-series work expensive and logistically difficult.
What the science actually claims, and what it does not
Kelp forests are among the most productive marine ecosystems on Earth. The blue-carbon argument rests on a specific export pathway: detached fronds or whole thalli sinking to deep-sea sediments where decomposition is slow enough that the carbon is effectively sequestered on century timescales. Published estimates of this export flux vary enormously, from negligible to potentially significant, and the fraction that reaches permanent burial rather than being remineralised in the water column remains poorly constrained.
No methodology approved by Verra, Gold Standard or any compliance market currently exists for kelp-based carbon credits. The Intergovernmental Panel on Climate Change's wetland supplement covers mangroves, tidal marshes and seagrasses but explicitly excludes macroalgae, citing insufficient data on net ecosystem carbon balance. Several voluntary methodology proposals are in development as of the mid-2020s, but none has cleared scientific peer review and registry approval. Anyone offering kelp carbon credits today is operating outside any recognised accounting framework. Satellite mapping is a necessary precondition for any future methodology, not a shortcut around the science.
Why near-infrared reflectance works, and where it fails
Kelp canopy at the water surface behaves spectrally more like terrestrial vegetation than like water. Macrocystis pyrifera and Ecklonia radiata both show strong chlorophyll-a absorption in the red (around 665 nm) and a characteristic reflectance rise in the near-infrared (700–900 nm) driven by cell-wall scattering. Water, by contrast, absorbs NIR almost completely. This contrast is the physical basis of kelp detection: a normalised difference vegetation index (NDVI) or a floating algae index applied to coastal imagery will separate surface canopy from open water with reasonable reliability in clear conditions.
The red-edge bands on Sentinel-2 (B5 at 705 nm, B6 at 740 nm, B7 at 783 nm) add discrimination power over standard NDVI, helping to separate kelp from other floating material such as Sargassum or surface foam. Published work using Sentinel-2 over Californian and South African kelp beds has achieved overall accuracies of 80–90 percent for canopy presence or absence at 10–20 m scale, with the main confusion being between sparse canopy and turbid water.
The method breaks down in several well-documented ways. Turbid water raises NIR-equivalent backscatter and produces false positives. Submerged canopy, which may constitute the majority of kelp biomass, is invisible: NIR penetrates only the top few centimetres of seawater. Persistent cloud cover over productive kelp regions (coastal California, southern Australia, South Africa, Patagonia) can reduce usable Sentinel-2 acquisitions to fewer than a dozen per year in some seasons. And the method is entirely blind to kelp density within the canopy, so extent maps cannot be directly converted to biomass without in-situ calibration.
Building a canopy map that a carbon auditor could use
A credible baseline map requires more than a single cloud-free image. The standard approach is to composite multiple acquisitions across a season, masking cloud and glint, then apply a spectral classifier trained on field-validated reference points. For Macrocystis, published spectral libraries and training datasets exist from work by the Santa Barbara Coastal LTER and from ESA-funded studies over South African Ecklonia beds. These provide a starting point, though local water-optical conditions require site-specific tuning.
Change detection over time demands consistent preprocessing: atmospheric correction to surface reflectance, sun-glint correction using the SWIR band (Sentinel-2 B11 or B12), and tidal-state normalisation where canopy extent varies with water level. Landsat's archive allows reconstruction of decadal baselines, which any additionality argument will eventually require. The honest limit is that archive-period classifications carry higher uncertainty than modern Sentinel-2 classifications, because Landsat's 30 m resolution misses patchy or narrow canopy features that would register in a 10 m image.
From extent to carbon: the gap that remote sensing cannot close alone
Satellite-derived canopy area is an input to carbon estimation, not an output. Converting area to carbon stock requires assumptions about canopy biomass density, the fraction of production exported below the remineralisation depth, and the permanence of that export. None of these parameters is currently measurable from orbit. Acoustic surveys, drift-tracking studies and sediment core analysis from the scientific literature provide ranges, but those ranges span an order of magnitude.
This is not a reason to avoid mapping. It is a reason to be precise about what the map delivers: a spatially explicit, time-stamped record of surface canopy extent that can be audited, repeated and compared across years. If and when a carbon methodology is approved, that record becomes the MRV backbone. Without it, any future accounting starts from scratch. The mapping investment is defensible now precisely because it is not contingent on the methodology debate resolving quickly.
What a practical programme looks like
A site-scale kelp monitoring programme would typically combine Sentinel-2 for systematic seasonal coverage, Planet SuperDove for high-resolution validation of classifier boundaries, and a Landsat retrospective for baseline setting. The analytical outputs are canopy extent polygons per season, a change-detection time series, and an uncertainty layer that flags pixels where turbidity or cloud limits confidence. Delivery as GeoTIFF and vector GIS layers is standard; integration with a carbon registry's project boundary shapefile is straightforward.
Satellize has built comparable multi-sensor coastal analytics pipelines, including spectral classification work under the Tonga crop-estimation programme, and can adapt that infrastructure to kelp-coast environments. The sensor physics is the same; the training data and water-optical corrections are site-specific. A scoping engagement would begin with an archive assessment of cloud-free coverage over the target coastline before any classification work is commissioned.
The honest commercial position: this is preparatory science, not credit issuance. Buyers should be research institutions, governments building national blue-carbon inventories, or project developers who want a defensible spatial baseline ready when methodology approval eventually arrives. Anyone expecting to sell credits within a two-year horizon is working on an unrealistic timeline.
Typical figures
| Best spatial resolution (canopy mapping) | 3 m (Planet SuperDove); 10 m (Sentinel-2 B4/B8); 30 m (Landsat 8/9) |
| Revisit frequency | Near-daily (Planet); 5 days (Sentinel-2, two satellites); 8 days (Landsat 8+9 combined) |
| Key spectral bands | Red (~665 nm), NIR (~842 nm), red-edge (705, 740, 783 nm on Sentinel-2 MSI) |
| Minimum detectable canopy patch | ~100 m² at 10 m resolution (Sentinel-2); ~20 m² at 3 m resolution (SuperDove); both require clear water and surface expression |
| Water-clarity constraint | Reliable detection requires Secchi depth >3 m; turbid estuarine or upwelling zones degrade accuracy significantly |
| Submerged canopy detection | Not possible with passive optical sensors; NIR penetration in seawater is centimetric |
| Cloud impact | Usable acquisitions can fall to <12 per year in persistently cloudy kelp regions; seasonal compositing required |
| Archive depth | 2015 to present (Sentinel-2); 1984 to present (Landsat 5/7/8/9); 2016 to present (Planet, commercial) |
| Classification accuracy (published range) | 80–90% overall accuracy for canopy presence/absence at 10–20 m scale in published Sentinel-2 studies over Macrocystis and Ecklonia beds |
| Delivery formats | GeoTIFF (classified canopy raster), vector polygon (GIS), seasonal change-detection report, uncertainty layer |
Analytics Satellize can run
| Seasonal canopy extent map | Spectral index classification (NDVI, floating algae index, red-edge ratio) on atmospherically corrected Sentinel-2 or Planet imagery, with cloud and glint masking | GeoTIFF and polygon shapefile of canopy presence per season, with per-pixel confidence score |
| Decadal baseline reconstruction | Landsat 5/7/8/9 archive time series with consistent surface-reflectance preprocessing; change-vector analysis to identify long-term canopy gain or loss | Annual canopy area time series (CSV + chart), polygon layers for selected reference years |
| Turbidity and water-quality mask | Suspended sediment index from Sentinel-2 visible and SWIR bands; flags acquisitions and pixels where optical water quality degrades classification reliability | Per-acquisition quality flag layer; summary of usable observation frequency per site per year |
| Canopy change alert | Bi-monthly difference between current and prior-season canopy extent polygon; threshold-based flagging of area losses exceeding a user-defined percentage | Email or API alert with change polygon and magnitude estimate |
| High-resolution boundary validation | Planet SuperDove or WorldView-3 classification at 3 m or sub-metre scale over selected transects, used to assess commission and omission errors in the 10 m Sentinel-2 product | Accuracy assessment report with confusion matrix and recommended confidence zones |
| Project boundary canopy inventory | Intersection of classified canopy raster with user-supplied carbon project boundary shapefile; area statistics per class per period | Tabular area summary and clipped GIS layer formatted for registry submission support |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.